Phương pháp cải tiến hệ thống nhận dạng biển giao thông bằng ứng dụng yolo=Method to improve the yolo-based traffic sign identification system

0 Người đánh giá. Xếp hạng trung bình 0

Tác giả: Thịnh Mạnh Đức Trần

Ngôn ngữ: vie

Ký hiệu phân loại:

Thông tin xuất bản: Khoa học và Công nghệ (ĐH Công nghiệp TP. HCM), 2024

Mô tả vật lý: tr.81-96

Bộ sưu tập: Metadata

ID: 245002

Thêm vào giỏ

Hệ thống biển báo giao thông được nhận dạng chính xác và nhanh chóng đã góp phần cải thiện việc quản lý giao thông trong thời đại chuyển đổi số theo chủ trương của chính phủ hiện nay. Hệ thống này đã được nghiên cứu và phát triển mạnh trong những năm gần đây với các giải thuật kinh điển về xử lý ảnh đến các giải thuật với ứng dụng trí tuệ nhân tạo mà thuật toán You Only Look Once (YOLO) dựa trên kỹ thuật thị giác máy tính là một đại diện đáng để quan tâm.Trong bài báo này, một phương pháp nhằm cải tiến hệ thống nhận dạng biển báo giao thông bằng ứng dụng YOLO như vậy được nghiên cứu về phương diện số lượng biển báo nhận dạng được, tốc độ nhận dạng và độ chính xác. Hệ thống được cải tiến để nhận dạng 9 loại biển báo giao thông khác nhau bao gồm: biến báo cấm rẻ trái, biển báo cấm rẻ phải, biển báo cấm rẽ trái-phải, biển báo cấm dừng-cấm đỗ, biển báo cấm đỗ, biển báo cấm ô tô rẽ phải, biển báo cấm ô tô rẽ trái, biển báo cấm quay đầu và biển báo cấm đi thẳng. Hệ thống được cải tiến và huấn luyện lại với các hình ảnh được thu thập ở các con đường trên thành phố Hồ Chí Minh bao gồm 735 ảnh biển báo cấm rẻ trái, 713 ảnh biển báo cấm rẻ phải, 177 ảnh cấm rẽ trái-phải, 752 ảnh biển báo cấm dừng-cấm đỗ, 629 ảnh biển báo cấm đỗ, 191 ảnh cấm ô tô rẽ phải, 143 ảnh cấm ô tô rẽ trái, 171 cấm quay đầu và 109 cấm đi thẳng. Hệ thống sau đó được kiểm thử thực nghiệm trên thực địa cho độ chính xác nhận dạng theo độ đo mAP@.5 lần lượt như sau: 99.4% nhận dạng đúng biển báo cấm rẽ trái, 99.3% nhận dạng đúng biển báo cấm rẽ phải, 95.6% nhận dạng đúng biển báo cấm dừng-cấm đỗ, 95.3% nhận dạng đúng biển cấm đỗ, 98.6% nhận dạng đúng biển cấm quay đầu, 93.7% nhận dạng đúng biển cấm ô tô rẽ phải, 94.5% nhận dạng đúng biển cấm ô tô rẽ trái, 99.5% nhận dạng đúng biển cấm rẽ trái-phải và cuối cùng là 93.4% nhận dạng đúng biển cấm đi thẳng.The system of accurately and quickly identified traffic signs has contributed to improving traffic management in the era of digital transformation according to the current government policy. This system has been researched and developed strongly in recent years from classic algorithms on image processing to algorithms with artificial intelligence applications such as the You Only Look Once (YOLO) algorithm based on the technology. Computer vision technique is a representative worthy of attention. In this article, a method to improve the traffic sign recognition system using the YOLO application is studied in terms of the number of recognized signs. format, recognition speed and accuracy. The system has been improved to recognize 9 different types of traffic signs including: no turning left sign, no turning right sign, no left-right turn sign, no stopping-no parking sign, no stopping sign. parking, signs prohibiting cars from turning right, signs prohibiting cars from turning left, signs prohibiting turning around, and signs prohibiting cars from going straight. The system was improved and retrained with images collected from roads in Ho Chi Minh City including 735 images of signs prohibiting left turns, 713 images of signs prohibiting right turns, 177 images prohibiting left turns - right, 752 photos of no-stop-no-parking signs, 629 photos of no-parking signs, 191 photos of prohibiting cars from turning right, 143 photos of prohibiting cars from turning left, 171 prohibiting turning around and 109 prohibiting going straight.

1. Đọc trực tuyến

1. Đọc trực tuyến

Tạo bộ sưu tập với mã QR